The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the user to establish persistent background execution for the data harvesting pipeline, which maintains access across sessions. Evidence includes:
Instructions to add daily recurring tasks to the system crontab (0 3 * * * cd ~/.personal-harvest && python3 ...).
A template for creating a macOS launchd service agent (com.personal-harvest.plist) that automatically executes the Python harvesting scripts at regular intervals.
[DATA_EXFILTRATION]: The skill accesses sensitive local application data and manages authenticated session state:
It programmatically reads the internal SQLite databases of the WeChat Reading application (~/Library/Containers/com.tencent.WeReadMac/Data/Library/Application Support/WeRead/*.db) to extract private reading history and notes.
It accesses the Kindle My Clippings.txt file for local data harvesting.
It utilizes persistent browser contexts to automate authenticated sessions and stores harvested session cookies locally in ~/.personal-harvest/cookies.json.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted data from multiple external sources which is then structured for use by the agent.
Ingestion points: Data is fetched from Douban (web), Bilibili (API), Xiaohongshu (web), Douyin (metadata resolver), and local files (Kindle text, WeChat Read SQLite DB).
Boundary markers: None; the skill does not wrap processed content in delimiters or include instructions to ignore embedded commands.
Capability inventory: The skill operates using high-privilege bash and computer tools.
Sanitization: None; the skill extracts raw text (titles, user notes, summaries) from platform DOM elements and local databases without filtering or validation.

personal-data-harvester